I recently graduated from the University of Toronto (UofT) - or as we like to call it, the “Harvard of the North.” It marked a very real ending to four long years of college.
UofT was intense. I met some of the most brilliant people I know there, and I’m deeply grateful that many of them took the time to attend my graduation. Leaving felt bittersweet.
Six months later, my life looks vastly different. I’d just wrapped up an AI research program at the New Turing Institute (NTI) by Dr. Thang Luong (Google DeepMind), and I’m about to start work as an AI Engineer. But the path wasn’t linear. Along the way, I tried on more hats than I ever expected: tutor, marketing associate, and even a hotel receptionist.
I’ve done some things right, but I also made a lot of mistakes.
In writing this blog, I hope that it serves as a reminder to myself of the biggest lessons I learned this year - and perhaps it may be helpful to some others too.
Play to your strength (not just your effort)
In university, I took pride in choosing the hardest math courses available. I believed real learning came from suffering – if it didn’t hurt, it didn’t count. So every semester, I pushed myself toward the most demanding classes.
Over time, I learned a humbling truth. I was good at math - but not exceptional in the way some of my peers were. In those rooms, I was competing on the same exams and metrics as people operating at a different level. No amount of extra effort changed the fact that the game rewarded a very specific kind of brilliance.
That forced a shift. If I wanted to truly excel, I needed to find a different axis - one where my strengths could actually compound. That axis turned out to be communication.
Looking back, I’d been training for this most of my life. I’d performed piano in front of hundreds of people to build confidence; I’d placed top five at a national TEDx competition; and many of the hackathons I won were decided not just by code, but by how the idea was sold.
When I joined NTI, I was surrounded by PhDs, elite CS students, and experienced engineers. I knew trying to out-compete them purely on technical depth wasn’t a winning strategy. So I leaned into what I did best.
I spent a lot of time thinking about how to make ideas legible to people outside my immediate niche. When presenting my work, I designed slides so that even a quick skim could convey the core ideas. During lectures with researchers from labs like Meta and Stanford, I prepared questions in advance to try to surface more insights for all audiences. I even found myself helping peers refine their presentations, knowing that this was a concrete way I could add value.
I still took my technical work seriously, but I also focused on the dimension where I had a natural advantage. I’m confident that I was more effective because of that choice.
Find your natural strengths, then deliberately place yourself in situations where they compound and move the needle
Useful > Impressive
This is a lesson I learned the hard way.
At NTI, my research focused on model routing, with the goal of reducing API costs for LLM systems. But if I’m being honest, much of my energy went into chasing something that looked innovative – something that would impress experts – rather than focusing on the core problem: building a deployable solution that actually lowers costs.
That bias had consequences. Instead of seriously exploring simpler, reliable approaches, I doubled down on a single “clever” idea. As the team lead, I pulled others down the same rabbit hole. When the idea didn’t work, our instinct wasn’t to step back, but rather to keep fixing the same idea, even as better alternatives clearly existed.
In the end, the strongest research projects weren’t always the most sophisticated ones. They were ones that solved a clear pain point and added real values. Ours, despite being “complete” on paper, couldn’t. We tried to re-invent the wheel when the wheel wasn’t broken.
I applied this lesson when building NeurIPS 2025 Map to explore AI research papers.
This time, I focused strictly on usefulness: fast search, clear navigation, and just enough information (ELI5 summaries) to help someone decide whether a paper was worth reading. I only worried about “impressiveness” at the very end, refining the UI to attract people to try it in the first place. Over 650 people have used it, and some of the most meaningful feedback I received wasn’t that it was impressive, but that it was useful.
Looking back, I was optimizing for looking impressive instead of being useful. While strong presentations matter, it can’t compensate for a lack of substance. You still have to walk the talk.
Start from the problem, not the clever solution; provide values to people
Go the extra mile
During my years at UofT, one of the people I learned the most from was Dr. Kennedy Obina Idu, my Multivariable Calculus instructor. Beyond being an exceptional educator, he is one of the wisest people I know.
He once shared his story about “beating the odds”, and one line he told me has stayed with me ever since. It went something like:
“To beat the odds, you must be willing to do what others won’t”
I realized this didn’t just mean working hard, but also taking on tasks that require extra effort without any guarantee of a payoff – the type of work most people skip because it feels “inefficient” or risky.
For a long time, I talked myself out of these things. I’d calculate the effort, see the uncertain outcome, and decide it wasn’t worth it. Yet, I always looked back with regret, thinking, I wish I’d just done it.
More recently, I stopped calculating and started doing. I built the NeurIPS 2025 Map out of my own frustration with navigating research papers. When I read a research paper I found interesting, I re-implemented the code and cold-messaged the author, a researcher at Thinking Machines, which led to feedback from the team.
Eventually, these moments of “unrequired” effort led to my current AI Engineer role and, to my surprise, even sparked interest from companies like Shopify.
The “extra mile” is rarely crowded, and it compounds faster than you think
Frankly, admitting your own misjudgements is never easy. But you live and you learn, and I’ve learned the most by owning up to mistakes and taking the time to reflect upon them.
Cheers to 2025, and may 2026 be a year of even greater learning!